[openib-general] ib_req_ncomp_notif in core_ layer

Michael S. Tsirkin mst at mellanox.co.il
Sun Aug 15 07:09:11 PDT 2004


Quoting r. Sean Hefty (mshefty at ichips.intel.com) "Re: [openib-general] ib_req_ncomp_notif in core_ layer":
> On Thu, 12 Aug 2004 09:47:18 +0300
> "Michael S. Tsirkin" <mst at mellanox.co.il> wrote:
> 
> > Either way this would have to be documented.
> > What do others here think?
> 
> Thanks for the clarification.
>  I think we want ib_req_notify_cq to behave as you described.  How does
>  the current implementation of req_ncomp_notify work?  What options does
>  the hardware support for n>1?

Not sure which implementation do you refer to.
Hardware, for n>1, reports event after arm N  if and only if

1. there was completion with error or solicited
   (or descriptor marked for event)
   after the last completion event, and this completion was
   not yet polled when arm was performed

  [ This is same as for request for solicited notification]

or

2.  At least n completions where not polled yet



(2) Means that if you did arm N and you see an event,
    you really want to do poll at least once or you'll get
    another event immediately.


> I will also add some documentation around these calls (depending on if
> they can be combined) for clarification.




More information about the general mailing list